RCDKSTAR

Introduction

RCDKSTAR stands for R-Chemistry Development Kit Structure-Activity Relationships. RCDKSTAR is a lightweight BioAssay aggregator and structural-activity relationship modeling suite.

Installation

shiny::runGitHub( "RCDKSTAR", "jeffrey-plume")

Modeling with RCDKSTAR

Select a protein coding gene from the drop-down list to populate the screen with summary data from PubChem BioAssays targeting the protein product. Filter the data as desired. Error and warning messages are displayed if something is off. BioAssays with the same target may have different endpoints, which can affect the activity outcome of the experiments. Choose a method for fingerprinting the molecules and any additional calculated properties. Hold CTRL to select multiple additional properties. Press ‘Load Properties’ to retrieve the desired values. Preprocessing of the data can be done by checking any of the checkbox options. Once satisfied, click “Train Model”. Depending on the complexity of the model and computing power, this may take a while. By Default, 75% of the curated dataset is used for training and the remainder for validation. Once complete, tables displaying summary statistics, rank-order motifs by importance, and probability outcomes of the remaining 25% of the data not used to train the model. Click “Test Unknowns” to apply the model to other compounds. If no SDF compound structural file is imported, compounds of the Drug Repurposing Hub by the Broad Institute are analyzed for activity.